Education
University
Higher School of Economics (HSE) Bachelor’s Programme 'Information Science and Computation Technology' Minor 'Applied Statistical Analysis' GPA 9.13 / 10 Sep 2018 - Oct 2021
Courses
| Open Machine Learning Course | Open Data Science mlcourse.ai |
| Introduction to Data Science and Machine Learning | Stepik, Bioinformatics Institute |
| Financial Markets | Coursera, Yale University |
| Algorithmic computation | Coursera, Higher School of Economics |
Experience
Research Engineer
ITMO, National Center for Cognitive Technologies Algoritmics of Complex Systems Saint Petersburg, Russia
| I managed to test all available motif extraction algorithms |
| Implemented Python wrapper for the state-of-art algorithm (GTscanner/GTriescanner) |
| Discovered advantages of Motif Significance Profile based on Z-scores for 4-5 motifs |
| Created Diffusion Dynamics Regression Model on Networks Using Sub-graph Motif Distribution |
Junior Data Scientist
VTB Bank Department of Data Analysis and Modelling Moscow, Russia
| Model of the client’s risk appetite as a part of recommendation system for investment products |
| Exploratory analysis of train dataset |
| Feature selection from Impala DWH |
| Data Cleaning |
Skills
| Programming languages | Python, SQL, C++, MATLAB |
| Tools | Sklearn, SciPy, Dask, Numpy, Networkx, Pandas, Matplotlib, Seaborn, PySpark, Hive, Impala |
| Algorithms | Boostings, Linear regression, Logit, Node2Vec, kNN, KMeans, SVD |
| Tasks | Recommended systems, Link Prediction, Uplift modelling, Anomaly detection |
| Languages | English [upper-intermediate], Russian [native] |
| Other | Git, Linux, Bash, LaTeX |
Projects
BIGTARGET Lenta & Microsoft 4th place Uplift modelling task in retail. Making an extra profit by communicating with customers I applied classical approach - target transformation + boosting, but with greedy algorithm of feature selection based on averaging of cross-validations. I managed to create a lightweight and stable model with using only CPU which is ready to meet production needs
FinNet Challenge Tochka Bank & ITMO 3rd place The task was to predict relevant partnership between bank clients I combined recommended systems and link prediction approaches and used SVD matrix factorization, Node2Vec embeddings and boosting of course :)
X5 Retail Hero Uplift Modeling 40th place - top 15% The goal is to target those customers who wouldn’t have made a purchase without communication I aggregated and extracted a set of features and utilize them for boosting
Speach Recognition University project DTW-based algorithm recognizes words from a limited dictionary Concept in few words: sliding frame, Mel-frequency cepstral coefficients, Fast Fourier transform, Dynamic Time Warping
Neural Network Bit Flipping Algorithm for SCList Polar Decoder University project We managed to find architecture of Neural Network that outperforms LSTM in bit flipping algorithm. It improves BER at 40% comparing to LSTM model.
Certificates
Gallery
